OpenAI’s “Own Goal”

And why it is becoming increasingly difficult to take them at their word

Elon Musk’s lawsuit against OpenAI may fail, or even get dismissed. But it’s already brought some important facts to light. OpenAI was so upset about the optics that they rushed out a statement, proclaiming ‘We are dedicated to the OpenAI mission and have pursued it every step of the way.”

That statement is dubious. The original mission, according to public statement and legal filings, was to be “unconstrained by a need to generate a financial return”, “not organized for … private gain”, “seek[ing] to open source technology where applicable.” Their opening statement to the world, December 15, 2015 was a statement of selfless beauty; had they stuck to it, I might still be in their fan club:

Repeating for emphasis: “Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world. We’ll freely collaborate with others across many institutions and expect to work with companies to research and deploy new technologies.”

Nowadays they have almost entirely abandoned open-source, and are among the least open AI companies anywhere. They are not freely collaborating, they are not publishing the details of their work. Almost nothing is known about the precise internal mechanisms of GPT-4 or what it has trained on. Rather than sharing with everyone, they are mostly sharing with Microsoft, which has an exclusive right to some aspects of the software and models. And 49% of OpenAI’s first couple hundred billion dollars in profits goes to them. Rather than helping all humanity OpenAI is (I would argue) effectively stealing intellectual property from artists, writers, etc to increase their profits.

Whatever their history with Elon Musk may be, it is not honest to claim that the mission has remained unchanged.

§

But none of that is exactly news.

What is news is an email to Musk that they foolishly made public yesterday, in their counterattack. In dealing with the above-mentioned inconvenient truths, OpenAI’s preferred narrative historically has been essentially to say (in my paraphrase) “we always intended to open source stuff but eventually, when GPT-2 came out in 2019, decided that it might be too dangerous to release”. (Never mind that they now sell a far more capable model at competitive prices.)

In effort to slap back at Musk they posted the email below from cofounder Ilya Sutskever to Musk, just three weeks after the beautiful initial mission statement.

Their point, which they were eager to make yesterday, is that they advised Musk that they wouldn’t be following the open-source mission, and that Musk acknowledged the above email. This may help them in court (though he had already publicly committed himself and might claim he was under duress).

But seen from another perspective, releasing that email was a massive fuckup—because it shows decisively that OpenAI knew for years that they had no intention of being as open as their name or public statements suggested.

And the story gets worse. After I posted the startling email above on X, someone else unearthed something from The New Yorker that is arguably even more damning, a description of a May 2016 meeting, when it had already become apparent that many people had come to believe that OpenAI meant Open-sourced AI for all:

“Please try not to correct that.”

#notconsistentlycandid.

They knew they were misleading people and failed to take an opportunity to correct. Very reminiscent of when Sam Altman told the Senate in May 2023 that he had no direct equity in OpenAI when he could clarified that he did have indirect equity.

§

The thing is OpenAI told the State of California that (a) it would be a “charitable organization” and (b) that it would open source its technology “for the public benefit when applicable.” The two facts that have come out make it look like they never really intended to follow their own charter.

Yet year after year they continue to file as a nonprofit, as recently as several months ago.

§

Leopards rarely change their spots. Just a few days ago OpenAI (among others) signed a big open letter promising “to use AI to improve people’s lives and unlock a better future”.

In a video that circulated this week (not sure when it was recorded) Altman spoke of using AI to remedy injustice.

What I am seeing is the opposite.

On Tuesday, we saw strong evidence of covert racism from LLMs revealed by Valentin Hofmann; no public response from OpenAI.

Today a Microsoft employee blew a whistle on violent and sexual content generated by their Designer product that is powered by OpenAI’s image synthesis software DALL-E, saying “he has warned MSFT … but … the company isn’t taking appropriate action.” According to CNBC, “The AI service has depicted demons and monsters alongside terminology related to abortion rights, teenagers with assault rifles, sexualized images of women in violent tableaus, and underage drinking and drug use” despite guardrails that presumably try to eliminate these sorts of things.

And the apparent copyright infringements that Reid Southen and I discussed in early January continue apace two months later, as I found in a quick experiment this morning:

§

Please forgive me if I no longer feel comfortable taking OpenAI at their word.

Positive efforts “to use AI to improve people’s lives and unlock a better future” would be nice. But casting a blind eye to covert racism, apparent copyright infringement, and violent and degrading outputs is not.

Gary Marcus is increasingly dismayed by the gaps he is seeing between Silicon Valley’s promises and what is actually being delivered.